Predicting In-Season Corn Grain Yield Using Optical Sensors
Round 1
Reviewer 1 Report
Review manuscript Agronomy-1942189, "Predicting in-season maize grain yield using optical sensors”.
This manuscript studies the use of optical sensors for predicting corn/maize grain yield.
This is a rather low quality paper, because lots of information is missing in the manuscript to make this a scientific sound manuscript. In addition, what has been described, for instance in terms of methods, is very confusing and scattered over introduction and methods sections without being really to the point describing the used methods.
I will list my main comments to this manuscript.
1. In the title “maize” is mentioned whereas the word “corn” is used throughout the manuscript. The authors should be consistent in their naming.
2. The authors should define the corn growth stages somewhere in the manuscript. Now they assume the readers know about the meaning of VT, R1, V10, etcetera. This needs to be introduced somewhere.
3. In the abstract (line 22) the authors state “SCCCI consistently outperformed other VIs in yield prediction” whereas this was only analysed for the Crop Circle and the MicaSense sensors. Moreover, the last sentence of the abstract is a very general statement that is much too general. You can state this always in such a study, so it is not a key outcome of this study and it is not of interest for the reader.
4. The introduction (section 1) is very weak. It should at least start with a paragraph describing why nitrogen is so important for yield prediction. Now this section starts with mentioning the crop yield goal (CYG) method, which is not used at all further on in the manuscript.
Line 38-39: The correlation between N uptake and VI is not that obvious and should be explained further in the introduction.
Line 40-46: Here various methods are introduced without explaining clearly what is meant. This should be put in more general wording here and details should be introduced in section 2. Now it raises a lot of questions. What are “raw VI values”? In section 2 it is described that the VIs are based on calibrated reflectances. So, they are not raw anymore. What is the effect of dividing the NDVI by GDD (called INSEY)? Is INSEY an estimated yield or (expressed as) a VI (line 41-42)? How is the response index (RI) derived (line 44)?
5. Acronyms should be written in full when you use them for the first time in the body text (like NDVI, NDRE, SPAD, etc.), even if they already have been used and explained in the abstract. The abstract stands separate from the body text in this respect.
6. Line 85: When and how were the yield data collected (line 86: “then”?)?
7. Line 90: DOI of reference 2 is not found (error message on 19-09-2022).
8. Line 138: Vegetation Indices
INSEY: what is the effect of NDVI saturation in this index?
NDVI saturation can be a serious problem. Why was an orthogonal type of index (difference VI) not tested, because it is known that these have less saturation effects and often perform better in yield prediction models.
9. The SPAD sensor mostly is used for measuring chlorophyll content at the leaf level. It looks like the 2 spectral bands are used in this study for determining vegetation indices. However, the authors are not very clear on this. For instance, what information (spectral indices) were used from SPAD in the VI comparisons (line 157-158)?
10. Line 173: Was an independent validation performed in this study? Without independent validation of the results (or a cross validation) the merit of such a study is limited.
11. Section 3 is an inferior quality section. The text now provides an overview of the most important numbers that can also be found in the tables. So, it mainly is duplication. The results section should be more descriptive in terms of the most important results by referring to the tables for the quantitative numbers.
Moreover the authors should clearly indicate what dates (or growth stages) have been used for the various subsections. It may be found in section 2, but then the reader has to dig again through that part of the manuscript. Now the reader will wonder why the MicaSense is performing moderately in table 4, whereas its results in table 6 are very good. This does not seem to be consistent.
12. Table 4 provides results of a comparison, but it is not clear with what the VI are compared to provide the given statistics in the last columns of the table. This also applies to subsequent tables.
13. There are no graphs at all in the manuscript. There is so much value of introducing a few graphs of the regressions for the main results. For instance, are relations indeed linear? How serious is the scatter (homoscedasticity)?
Author Response
In the title “maize” is mentioned whereas the word “corn” is used throughout the manuscript. The authors should be consistent in their naming.
Agreed. The title has been updated to corn instead of maize. Thank you
The authors should define the corn growth stages somewhere in the manuscript. Now they assume the readers know about the meaning of VT, R1, V10, etcetera. This needs to be introduced somewhere.
Thank you for your input. A section delineating required information related to the staging process is now included.
In the abstract (line 22) the authors state “SCCCI consistently outperformed other VIs in yield prediction” whereas this was only analyzed for the Crop Circle and the MicaSense sensors.
Very true, SCCCI outperformed other VIs and only Crop Circle and MicaSense have the capability to compute SCCCI.
Moreover, the last sentence of the abstract is a very general statement that is much too general. You can state this always in such a study, so it is not a key outcome of this study and it is not of interest for the reader.
Agreed. The abstract is updated as suggested and key findings from this study are now incorporated.
The introduction (section 1) is very weak. It should at least start with a paragraph describing why nitrogen is so important for yield prediction. Now this section starts with mentioning the crop yield goal (CYG) method, which is not used at all further on in the manuscript.
Agreed. A paragraph highlighting the role N plays in corn production is now included.
Line 38-39: The correlation between N uptake and VI is not that obvious and should be explained further in the introduction.
Agreed. This is now further elaborated in the introduction.
Line 40-46: Here various methods are introduced without explaining clearly what is meant. This should be put in more general wording here and details should be introduced in section 2. Now it raises a lot of questions. What are “raw VI values”? In section 2 it is described that the VIs are based on calibrated reflectances. So, they are not raw anymore. What is the effect of dividing the NDVI by GDD (called INSEY)? Is INSEY an estimated yield or (expressed as) a VI (line 41-42)? How is the response index (RI) derived (line 44)?
The purpose of this section was to give a general background and use it for wider context of current sensor-based N management decisions. Some characteristics, such as the RI, are not going to be as detailed as it is only there to explain how some methods are used.
You are completely right about the raw values being misleading, we didn’t consider how those would be interpreted. A clarification within the text is now included when these are not truly raw values such as with MicaSense data. INSEY is a modified VI measurement and allows seasonal weather effects and planting date to be accounted for within the VI reading (Raun, et al., 2002). Furthermore, a description on calculation of relative values is also included.
Acronyms should be written in full when you use them for the first time in the body text (like NDVI, NDRE, SPAD, etc.), even if they already have been used and explained in the abstract. The abstract stands separate from the body text in this respect.
Agreed. The text has been updated to reflect this.
Line 85: When and how were the yield data collected (line 86: “then”?)?
The harvest date range for 2020 and 2021 and further information of how yield was collected within the text is now included. Thank you.
Line 90: DOI of reference 2 is not found (error message on 19-09-2022).
This is an accepted paper in agronomy journal but final version is not published yet. Here is the link again: https://doi.org/10.1002/agj2.21179
Line 138: Vegetation Indices
INSEY: what is the effect of NDVI saturation in this index?
NDVI saturation can be a serious problem. Why was an orthogonal type of index (difference VI) not tested, because it is known that these have less saturation effects and often perform better in yield prediction models.
INSEY corrects for days when growth is not possible and has been a better predictor for yield than the NDVI. It does compensate for saturation effects of NDVI up to an extent. The reason we used NDVI is because it is still the most common VI used for yield prediction in corn regardless of the issues you rightly highlighted. We used off the shelf sensors available and calculated other VIs such as NDRE and SCCCI that don’t saturate. The proximal sensors we used only have two to three bands and has limited VI capabilities compared to MicaSense with 5 bands which could result in many VI calculations.
The SPAD sensor mostly is used for measuring chlorophyll content at the leaf level. It looks like the 2 spectral bands are used in this study for determining vegetation indices. However, the authors are not very clear on this. For instance, what information (spectral indices) were used from SPAD in the VI comparisons (line 157-158)?
Correct. The SPAD index itself was used and compared with other VIs.
Line 173: Was an independent validation performed in this study? Without independent validation of the results (or a cross validation) the merit of such a study is limited.
At this stage of our work we just tested the feasibility of different sensors, VIs, and stages and where these sensors work best. Future research will solidify these finding with additional data and prediction models will be constructed and validated on independent dataset before suggesting further use.
Section 3 is an inferior quality section. The text now provides an overview of the most important numbers that can also be found in the tables. So, it mainly is duplication. The results section should be more descriptive in terms of the most important results by referring to the tables for the quantitative numbers.
The results section highlights the most prevalent results so that the reader can understand what is presented in the tables. However, this section is further improved as per your suggestions.
Moreover, the authors should clearly indicate what dates (or growth stages) have been used for the various subsections. It may be found in section 2, but then the reader has to dig again through that part of the manuscript. Now the reader will wonder why the MicaSense is performing moderately in table 4, whereas its results in table 6 are very good. This does not seem to be consistent.
Detail is now provided on the growth stages as requested. Thank you.
Table 4 provides results of a comparison, but it is not clear with what the VI are compared to provide the given statistics in the last columns of the table. This also applies to subsequent tables.
We are not exactly following what the reviewer is referring to and need further clarification. Each table caption defines the content of the table, for example, in table 4, the comparison is within the sensors where different methods of yield prediction by specific sensor type are compared.
There are no graphs at all in the manuscript. There is so much value of introducing a few graphs of the regressions for the main results. For instance, are relations indeed linear? How serious is the scatter (homoscedasticity)?
We completely agree with you. We heavily considered adding figures but were hesitant not too add duplication of results presented in tables. Two figures are now added to accompany the tables. Thank you.
Reviewer 2 Report
The manuscript “Predicting in-season maize grain yield using optical sensors” (agronomy-1942189), demonstrated the comparative sensors applied, vegetation index (VIs) by predicted and prevision yield in corn productions.
In general, The authors have done a good work, in Introduction sections. However, corrections were necessary in the Materials and Methods, Results and especially, Discussion sections. This article, which demonstrated expensive and relevant results and discussion, but minor recent literature was cited. The manuscript need major revision. In addition, English changes are required BY Native English experts its necessary to many correct phrases. Many phrases are unclear.
#Please, observe and standardize the terms throughout in a manuscript.
Answer:
- “Why only limited to yield corn in 2020 to 2021? “Field experiments at eight-site years were 16 established in Mississippi”.
-L161-164. Why was remove this data? I think your model based only 2020? Its not unclear for me. Could you explain, please?
-What is data of yield? That’s data are just as important as sensor, method of analysis and VI comparisons! Please, show them.
-Matherial and methods. Soil analysis?
-Discussion is not adequate and unclear. Does not discuss the data properly. Needs extensive revisions.
-L251-254. What is models? It is unclear. Was not the idea to perform comparisons between sensors, VIs and estimate the best models, combining VIs? Why was not a new VIs "created" and demonstrated using different sensors? Please, rephrase.
-Discussion. What are the main limitations for large-scale corn cultivation, production and application to prediction based on VIs in agriculture and other cultivated plants? Why your models are not good R2 <0.6?
-Explain and rephrase. Why “MicaSense best predicted yield at VT-R1 stage (R2 = 0.78-0.83)” was best to predicted? Perhaps perhaps this is the strong point of the work. However, there was no further discussion on the subject.
-#01: There is a scope for improvement in the discussion section: a) additional emphasis on the significance of the study, b) scientific contribution of the paper; c) perspectively to other plants to agronomic interestly;
-Conclusion: What are the future perspectives of Vis and sensor tests in manuscript by applied to improve other agronomic plants?
Minor points:
-Title: only the first capital letter of each word;
-Alphabetic order keywords;
-Table 01 – Why application rate was different between 2020 and 2021?
-Subtitle should be in topics, e.g.line 47, 92, 128, and other subtitle. Please check all manuscript.
L89. What is this? Its not good. Describe.
L94-96. Please. All standardization of nomenclature equipment/reagents when necessary. Example: Fabricant, City, State, Country (three-letter). Check all manuscript.
Table 4. Please, reframe. This is table and not frame. In addition, what is Y? Can you obtained that’s models?
Table 5. Please, reframe. This is table and not frame.
Table 6. Please, reframe. This is table and not frame.
Best Regards
Author Response
-“Why only limited to yield corn in 2020 to 2021? “Field experiments at eight-site years were 16 established in Mississippi”.
This line was removed from the paper. Within this paper, 8 site years would be 4 sites over the 2 years of this study.
-L161-164. Why was remove this data? I think your model based only 2020? Its not unclear for me. Could you explain, please?
For this section, we were trying to make the sensor data comparison as comparative as possible. To do this we chose the VT stage, which was most prominent across sensors. But since not every sensor had VT data in the 2021 year, it was best to remove the 2021 so that each sensor could be compared more equally.
-What is data of yield? That’s data are just as important as sensor, method of analysis and VI comparisons! Please, show them.
We have added additional information explaining how yield data was captured for this study. We have also added figures that contain yield data so that yield trends can be seen across the data sets. Thank you for your suggestion.
-Material and methods. Soil analysis?
Soil analysis information and results were added. Thank you.
-L251-254. What is models? It is unclear. Was not the idea to perform comparisons between sensors, VIs and estimate the best models, combining VIs? Why was not a new VIs "created" and demonstrated using different sensors? Please, rephrase.
A model in prediction research is the algorithm derived from plotting regressor and regressand variables. The purpose of this study was not to create additional VIs, but to provide a comprehensive examination on factors important for future algorithm creation. We particularly focused on the VI manipulation method. Our ability to create VIs in this study would be limited due to the Crop Circle and particularly the SPAD and GreenSeeker sensors only having a small amount of band combinations to test.
-Discussion. What are the main limitations for large-scale corn cultivation, production and application to prediction based on VIs in agriculture and other cultivated plants? Why your models are not good R2 <0.6?
That is a good question. Within the paper we mentioned how the correlation between yield and relative VI improves as the corn matures. This could be considered a limitation as there is a need for specialized equipment for N application at later stages yet that is when yield and VI is most correlated. As for the low R2, this is an aspect inherent to the study where there are multiple factors that are impacting yield variability. The goal is to see which method, sensor, VI, and stage combination best accounts for that variability from those multiple factors. When it is low (R2 <0.6) this combination was not successful in capturing that variability.
-Explain and rephrase. Why “MicaSense best predicted yield at VT-R1 stage (R2 = 0.78-0.83)” was best to predicted? Perhaps perhaps this is the strong point of the work. However, there was no further discussion on the subject.
Agreed. We have updated the discussion section and abstract to place a greater emphasis on the MicaSense’s relative SCCCI VT/R1 stage combination having the highest correlation.
-#01: There is a scope for improvement in the discussion section: a) additional emphasis on the significance of the study, b) scientific contribution of the paper; c) prospectively to other plants to agronomic interest;
We have placed a greater focus on the main point to be emphasized, that the Mica Sense sensor using relative SCCCI at later growth stages was best suited for grain yield prediction.
-Conclusion: What are the future perspectives of Vis and sensor tests in manuscript by applied to improve other agronomic plants?
While model capability in other crops was not part of our research aims as the ultimate goal is to create a corn N algorithm in Mississippi, similar research has been completed in other crops (Raper & Varco, 2014).
-Title: only the first capital letter of each word;
The title has been updated and each first letter is capitalized. Thank you.
-Alphabetic order keywords;
This has been changed. Thank you.
-Table 01 – Why application rate was different between 2020 and 2021?
Our study design changed for the sake of an additional research aim to test the difference in single or split application but should not have significantly affected the modeling of this study.
-Subtitle should be in topics, e.g.line 47, 92, 128, and other subtitle. Please check all manuscript.
We are not exactly following what the reviewer is referring to and need further clarification. These are currently labeled as sublevel headings.
L89. What is this? Its not good. Describe.
Agreed. This sentence was removed.
L94-96. Please. All standardization of nomenclature equipment/reagents when necessary. Example: Fabricant, City, State, Country (three-letter). Check all manuscript.
This has been updated so equipment location citation has been standardized.
Table 4. Please, reframe. This is table and not frame. In addition, what is Y? Can you obtained that’s models?
Y is a mathematical shorthand to refer to the equation created from plotting the data. The table has been reframed to match existing tables above.
Table 5. Please, reframe. This is table and not frame.
The table has been reframed to match existing tables above.
Table 6. Please, reframe. This is table and not frame.
The table has been reframed to match existing tables above. Thank you.
Round 2
Reviewer 1 Report
The quality of the manuscript has been improved significantly. Although this still is not the highest quality paper it will now be acceptable for publication after some minor additions/changes.
Here are my suggestions for improvement, before the manuscript is ready for publication:
1. Since no independent validation has been performed, I suggest to add a remark on this at the end of section 2 (after line 201 in the clean version), e.g.
“Since this study should be considered a feasibility study, no independent validation of results was performed. This has to be done in future research.”
2. At line 206 I suggest to add the following sentence:
“Best results for the Crop Circle and the MicaSense sensor are illustrated in Figure 1.”
3. In the caption of Figure 1 it should be stated that the graphs refer to “relative VIs”.
4. In the caption of Table 5 it should be stated that the table deals with “comparison in yield prediction”.
5. With the addition of Figure 2, Table 6 provides exactly the same information and should be left out.
6. The reference list at the end of the manuscript still needs concise editing. Several references still miss the name of the journal, volume, etcetera (e.g., refs 2, 7, 17, 21 (MSc thesis) and 22). Reference 1 provides the title twice. Also the use of italics should be made consistent.
Author Response
Since no independent validation has been performed, I suggest to add a remark on this at the end of section 2 (after line 201 in the clean version), e.g.
“Since this study should be considered a feasibility study, no independent validation of results was performed. This has to be done in future research.”
This has been added at the end of section 2. Thank you.
At line 206 I suggest to add the following sentence:
“Best results for the Crop Circle and the MicaSense sensor are illustrated in Figure 1.”
This has been added at line 206. Thank you.
In the caption of Figure 1 it should be stated that the graphs refer to “relative VIs”.
This has been clarified in Figure 1. Thank you.
In the caption of Table 5 it should be stated that the table deals with “comparison in yield prediction”.
This has been added in the caption of Table 5. Thank you.
With the addition of Figure 2, Table 6 provides exactly the same information and should be left out.
With your permission, we would like to keep Table 6. Table 6 does contain unique information with the RMSE values with the primary function of the addition of figures to provide a visual aid and give an indication of yield values.
The reference list at the end of the manuscript still needs concise editing. Several references still miss the name of the journal, volume, etcetera (e.g., refs 2, 7, 17, 21 (MSc thesis) and 22). Reference 1 provides the title twice. Also the use of italics should be made consistent.
The reference section has been updated for the entries you have mentioned along with others. Thank you.
Reviewer 2 Report
Dear Authors, I would like to thank the authors for addressing my comments. I consider the authors made important changes in the manuscript and it was highly improved. I recommend the publication of the manuscript in its current form. Best regards.
Author Response
Dear Authors, I would like to thank the authors for addressing my comments. I consider the authors made important changes in the manuscript and it was highly improved. I recommend the publication of the manuscript in its current form. Best regards.
Thank you for your helpful recommendations for the paper. Best regards.
